The second method (Fig. 1B) relaxes the assumption that the base calls and the alignments are correct and instead considers two types of uncertainty related to determining ai and Ni , namely the uncertainty encoded in the base call qij∈[0, 1] which represents the probability that the stated base is correct for read j∈(1, …, Ni) at position i; and rij∈[0, 1] representing the probability that read j aligns to its stated position in the genome. Note that although mapping quality is derived in part from base qualities, considering these quantities as independent allows us to encode the fact that base qualities are position specific, while mapping qualities are constant for all bases in the read. The input data for this method can be visualized as shown in Figure 1B: high mapping quality is shown as dark background and high base quality as bright foreground, high contrast positions indicate positions where the data are more trustworthy. We show in Section 2.4 how to explicitly model these uncertainties to perform soft probabilistic weighting of the data rather than thresholding the uncertainties
